Cross-country multi-modal evidence links Aspergillus to biliary atresia

Prompted by a clinical case in which antifungal therapy improved neonatal cholangitis, we investigated a potential association between gut fungal colonization, particularly by Aspergillus spp., and the development of BA in infants. We employed a comprehensive approach that integrated clinical case reviews, molecular sequencing, and validation using a publicly available fecal metagenomic dataset. Clinical samples were analyzed from two hospitals in Taiwan: National Taiwan University Children’s Hospital and En Chu Kong Hospital. Participant demographics and diagnostic information are summarized in Table 1. To further support our findings and assess environmental fungal exposure, we performed epidemiological analyses across Taiwan, the United Kingdom, and Japan.

Detailed protocols for fecal DNA extraction, 16S/18S PCR, Sanger sequencing, and TA cloning are provided in the Supplementary Methods (Additional file 1).

Sanger sequencing results were processed in SnapGene (Dotmatics, Boston, MA, USA; www.snapgene.com) for visualization and editing. Taxonomic identities were assigned through nucleotide BLAST [9] searches against the NCBI nr/nt database [10] using the NCBI web server. Data visualization was performed in the Python library seaborn [11]. For fungal taxonomic profiling, Kraken 2 [12] was employed. From each sample, 2 million sequence reads were randomly selected, and analyses were repeated five times. The averaged output from these replicates, totaling 10 million reads, was used to represent each sample. Forward and reverse reads were processed separately to ensure consistency and coverage.

To estimate case numbers from published figures, Fig. 1 in Sung et al. [13] was digitally enlarged by approximately 330%, establishing a visual scale in which 5 cm represented 300 cases of Aspergillosis and 4 cm represented 100 cases of Candidiasis. The vertical heights of plotted data points, corresponding to monthly data from March to June across 2018–2021, were measured manually with a ruler. Similarly, incidence estimates for Aspergillus spp. were derived from Fig. 1 in Hsiue et al. [14], where 1 mm equaled 1 case per 1,000 hospital admissions and 5.3 cm represented 20% of total cancer admissions. To ensure reproducibility and minimize measurement bias, manual measurements were independently repeated by five additional individuals, including three non-authors, each using randomly adjusted image scales. Correlation coefficients showed excellent consistency across raters: UK-Aspergillus (mean r = 0.985, standard deviation [SD] = 0.002, coefficient of variation [CV] = 0.2%), UK-Candida (mean r = 0.55, SD = 0.028, CV = 5.2%), and TW-Aspergillus (mean r = 0.77, SD = 0.019, CV = 2.5%). Here, SD represents the dispersion of values around the mean, and CV (SD ÷ mean × 100%) indicates relative variability. These results demonstrate that the manual measurement approach is highly reproducible (Fig 3 and Figure S2, Additional file 1). For Japan, incidence data were extracted from published tables by Kume et al. [15] and Suzuki et al. [16], and directly retrieved from the Japanese Biliary Atresia Registry (https://jbas.net/registration-historicaldata).

Bild vergrößern

Statistical analyses were performed to assess associations and differences between groups. Fisher’s exact test was used to evaluate the presence of non-commensal fungi uniquely detected in cases compared with controls. Correlation analyses were conducted using the pearsonr function in SciPy [17], with Pearson’s r, 95% confidence intervals, and p-values reported. Regression plots display analytically calculated 95% confidence bands around the fitted line. Differences in fungal genus abundance between groups were assessed using Student’s t-test (two-sample, equal variances), with mean differences (Δ), 95% confidence intervals, and p-values reported. A two-sided p < 0.05 was considered statistically significant. All analyses and visualizations were performed in Python using NumPy, SciPy, pandas, seaborn, and matplotlib.

No a priori sample size or power calculations were performed due to the opportunistic, retrospective design. To guide prospective validation, we estimated required sample sizes from both observed and conservative effect sizes. For prevalence analyses of non-commensal fungi, planning around an absolute difference of 20–30% between BA and controls suggested that approximately 12–19 participants per group would be sufficient at 80% power and α = 0.05 (two-sample proportion test, assuming near-zero prevalence in controls). On a more conservative basis, allowing for a small background prevalence in controls and other planning safeguards, larger samples of approximately 22–36 per group may be advisable. For correlation analyses of annual BA incidence against Aspergillus burden, the observed effects (r ≈ 0.78–0.98 across countries) implied that ~8–11 annual observations would be sufficient for 80% power. Using a conservative target based on two standard deviations below the cross-country mean (r ≈ 0.67) required ~16 observations, while planning for a more moderate correlation (r ≈ 0.5) would necessitate ~30 observations. These estimates are intended to guide prospective confirmation and do not alter the exploratory nature of the present work.

A male newborn was delivered via cesarean section at 39 weeks of gestation due to fetal distress. The infant experienced cardiac arrest and exhibited thick meconium staining at birth but was successfully resuscitated and transferred to the Neonatal Intensive Care Unit. Severe hypoglycemia (blood glucose < 2 mg/dL) was detected on admission. Continuous infusion of 10% glucose via peripheral lines only barely maintained blood glucose levels. Because central lines were unavailable, early oral feeding with 50% concentrated glucose solution was initiated on day 1 (D1) and continued for one week. Insulin and cortisol responses were within normal ranges on D3, and glucose levels were mostly maintained above 40 mg/dL thereafter, with exceptions on D8 (14 mg/dL) and D25 (39 mg/dL). No metabolic acidosis was observed. Alanine aminotransferase was elevated on D1 (389 IU/L) but returned to near-normal levels by D8 (44 IU/L). The patient’s vital signs remained stable, and no neurological complications were noted.

However, ileus developed later in spite of full antibiotic coverage. Severe thrombocytopenia was observed on D8 and D11, while white blood cell (WBC) counts remained normal (Fig. 1A). A fungal infection was suspected, and intravenous (IV) fluconazole was initiated on D11. Clinical improvement was immediate, and fluconazole was discontinued on D19. However, leukocytosis developed (20,400/μL on D15; 28,300/μL on D19), prompting another course of IV antibiotics (D19–D26), which did not improve clinical parameters (D27, Fig. 1A).

During this period, eosinophilia (D27–D30; shades of WBC, Fig. 1A) and direct hyperbilirubinemia with elevated gamma-glutamyltransferase (γ-GT) were noted (D27, Fig. 1B). The patient also developed perianal dermatophytosis that responded to topical clotrimazole, along with the presence of light-colored stool (D26, Fig. 1C). Suspecting fungal cholangitis, IV amphotericin B was given from D27 to D29 (Fig. 1B). Because bilirubin levels did not improve, antifungal therapy was switched to oral fluconazole on D30 (Fig. 1B). Interestingly, stool color normalized the following day (D31, Fig. 1C), with sustained improvement thereafter (D38, Fig. 1C). Hyperbilirubinemia gradually resolved (Fig. 1B), and abdominal ultrasound revealed a well-distended gallbladder without bile duct obstruction (D40, Fig. 1C).

The patient was discharged on D41 while continuing oral fluconazole. Follow-up monitoring of blood counts, bilirubin, and γ-GT values remained normal (D55–D108, Fig. 1A–B). Oral fluconazole was tapered and discontinued. At two and a half years of age, the most recent evaluation showed that the child had normal growth and development.

Fungal cultures of blood and stool samples collected during hospitalization were all negative. We used fungal ribosomal 18S PCR [18, 19] to assess suspected fungal colonization from fecal DNA, with bacterial 16S V1–V9 PCR [19] as an internal control (Figure S1A, Additional File 1). A clear 18S V1–V8 band was amplified from the D28 fecal sample (Figure S1A, Additional File 1), despite IV amphotericin B administration from D27 to D29 (Fig. 1B). This V1–V8 band was no longer detectable on D31 and D38 following the start of oral fluconazole on D30 (Fig. 1B). PCR products from the V1–V8 region were further amplified for V7–V8 taxonomic classification by semi-nested PCR (red arrow, Figure S1A, Additional File 1). Sequencing identified Aspergillus spp. by nucleotide BLAST against the nr/nt database (Table 1; Figure S1B, Additional File 1).

Aspergillus spp. is not considered a typical commensal fungus of the gastrointestinal tract [20]. We hypothesized that BA, a disease often preceded by obstructive cholangitis, might also be associated with colonization by unusual fungi in the bowel. To test this, we examined archived fecal DNA from five BA cases and four healthy controls using 18S V1–V8 and semi-nested V7–V8 PCR (Figure S1C, Additional File 1). Four samples (BA01, BA03, BA05, and Control03) yielded positive results (Table 1; Figure S1C, red arrow, Additional File 1). Sequences from BA01 mapped to Cerrena spp. (Figure S1D, Additional File 1), while Aspergillus spp. was also identified in BA05. Control03 and BA03 were associated with Candida. Because both the cholangitis case and BA05 were linked to bowel colonization by Aspergillus spp., we aligned their 18S V7–V8 sequences. At least two loci showed distinct T vs. C nucleotide polymorphisms (red arrow, Figure S1E, Additional File 1), suggesting that the two patients were colonized by different Aspergillus species.

We expanded our investigation of fecal 18S sequences using semi-nested V4–V8 PCR. Five additional BA samples (BA06–BA10, Table 1), four normal controls (Control05–Control08, Table 1), and five cholestatic disease controls (Control09–Control13, Table 1) were included. Due to sequence heterogeneity in BA06, PCR products were cloned before Sanger sequencing. BLAST analysis revealed multiple fungi in BA06, including Aspergillus spp. (BA06, Table 1). Using the longer sequence span from V4–V8 PCR, the D28 sample from the index cholangitis patient mapped to Penicillium spp. (D28, Table 1), which is taxonomically related to Aspergillus spp. [21]. Among the four additional normal controls and five cholestatic controls, three yielded sequences from Candida and three from Malassezia (Control05–Control13, Table 1). In summary, three of ten BA samples contained non-commensal fungi, whereas none of the controls did (Table 1). Patients with BA or cholangitis were therefore significantly more likely to harbor non-commensal fungi than controls (Fisher’s exact test, p = 0.0311), with Aspergillus spp. detected repeatedly.

Building on these findings, we anticipated an expansion of Aspergillus populations, particularly in older BA cases. To validate this, we analyzed a publicly available dataset of fecal shotgun metagenomes from Song et al. [22], which revealed an unusual enrichment of Aspergillus in BA. The dataset comprised 16 BA cases and 10 matched controls, with half of the BA patients having undergone Kasai portoenterostomy. For each sample, two million forward reads were randomly selected for fungal taxonomic classification using Kraken 2 [12]. Five replicate subsamples were analyzed, totaling ten million reads per sample, and the average genus-level fungal abundance was used for comparisons.

In the forward reads, total fungal counts were comparable (p = 0.672) between BA (3543.5 ± 828.7 per two million reads, standard error [SE]) and controls (2983.6 ± 987.3 per two million reads, SE). Five genera were significantly more abundant in BA: Aspergillus, Kluyveromyces, Marasmius, Metarhizium, and Lachancea (Fig. 2A). In the reverse reads, four of these genera overlapped (Aspergillus, Kluyveromyces, Marasmius, and Metarhizium) (Fig. 2B). In both analyses, Aspergillus showed the largest mean difference between BA and controls, with nearly identical abundance distributions across read directions. For each comparison, mean differences (Δ) and 95% confidence intervals are displayed in the figure.

Bild vergrößern

The exclusive detection of Aspergillus spp. in BA and cholangitis cases (Table 1), together with confirmation from an independent dataset (Fig 2), led us to hypothesize a potential role for this fungus as a trigger in BA development. A recent UK report indicated that BA presentations declined during COVID-19 lockdowns [23], while another reported a simultaneous decrease in hospitalized patients with aspergillosis [13]. This parallel observation suggests that Aspergillus spp. may represent a shared environmental trigger for both conditions. Accordingly, a strong correlation between BA incidence and hospitalizations for aspergillosis would be anticipated.

The first UK lockdown lasted from March to June 2020, prompting us to analyze patient numbers during the corresponding months of 2018–2021. Using data reported by Sung et al. [13], we estimated patient numbers for Aspergillosis and Candidiasis during these periods and compared them with January–June BA case reports from Arshad et al. [23]. Our analysis revealed a strong correlation between BA cases and hospitalizations for Aspergillosis (r = 0.98, p = 0.02; Fig. 3A), supporting our hypothesis of a shared environmental trigger. In contrast, no significant correlation was observed with Candidiasis (r = 0.60, p = 0.40; Fig. 3B).

Bild vergrößern

In Taiwan, the annual incidence of BA from 2000 to 2009 [24, 25] closely paralleled the annual incidence of Aspergillus-positive isolates among hospital admissions adjusted for cancer-related cases [14], serving as a proxy for environmental fungal burden (r = 0.78, p = 0.01; Fig. 3C). In Japan, BA cases reported by the Biliary Atresia Registry [26] over 25 years strongly correlated with annual autopsy reports of visceral Aspergillosis [15, 16] (r = 0.85, p = 0.01; Fig. 3D), reflecting the community burden of Aspergillus. In contrast, no significant correlations were found with visceral Candidiasis (r = 0.54, p = 0.17; Fig. 3E), Cryptococcus (r = 0.42, p = 0.30; Fig. 3F), or Mucormycetes (r = 0.41, p = 0.31; Fig. 3G). For all correlations, the figures display both the 95% confidence intervals for Pearson’s r and shaded 95% confidence bands around the regression lines. These data underscore the distinct association of Aspergillus spp. in BA incidence.